Vernon Hills student draws inspiration from mom’s cancer battle to develop early detection tool

Aarnav Garg is putting artificial intelligence to use to help solve a human problem with a personal connection.

The Vernon Hills High School junior’s mother was diagnosed with Stage IV ovarian cancer when he was 10 years old. She is now cancer-free, and Aarnav, 16, has channeled his family's experience into creating an AI-powered early detection tool designed to support ovarian cancer diagnoses.

Garg trained his algorithm, which he calls OvaScan, on a public repository of ultrasound imaging. His prototype can identify tumors from a scan and provide insight into the tumor’s specific characteristics, including whether it is benign or malignant.

OvaScan currently exists as a mobile app, which Garg views as a jumping-off point to his larger vision of translating the prototype into a web application that can be directly integrated into an ultrasound machine.

In September, Garg was among about 40 high school students from across the country recognized as a Fellow by the Davidson Institute, which awards scholarships to “high-achieving students” who have completed a “significant piece of work,” according to the institute's website.

He also was named the 2026 Illinois BioGENEius Champion in the spring, beating out 32 other high school aged semifinalists in the statewide competition for independent science research. That award came with a $1,500 cash prize.

Garg cited OvaScan’s accessible user interface and low cost as what separates it from other AI-powered cancer-detection models.

“A machine learning model isn't that easy to interpret by clinicians and researchers and people who are actually working in the field, because they're not computer scientists,” Garg said. “The entire purpose of this project was to make all of that more accessible for those people that actually use it.”

He envisions OvaScan as being especially useful in under-resourced communities that might not have access to expert radiologists and struggle to afford other tools.

The American Cancer Society estimates that about 21,010 women will receive a diagnosis of ovarian cancer in 2026, and it lists ovarian cancer as the sixth-leading cause of cancer death among women.

Early detection is key to treatment.

A 2018 study from the Hematology/Oncology Clinics of North America estimated that Stage I ovarian cancers, which refer to tumors confined in the ovaries, carry a five-year survival rate of 90%. For Stage III and IV, a designation applied when the cancer spreads beyond the pelvis, the five-year survival rate drops to below 20%. That same 2018 study estimated that only 20% of ovarian cancer cases are diagnosed in Stages I and II.

Garg recalls reading similar statistics when his mom was diagnosed.

“I was old enough to know that this was a very serious thing,” he said. “My mom was just very positive about it as a whole despite everything. Her resilience really stuck with me.”

Garg worked with his mother to find the data he used to train his model. She reached out to her gynecologist’s office to see if they could provide any data — which wasn’t ultimately successful due to confidentiality issues — but eventually helped him find the public repository that became the foundation for his project.

“When I first told my mom about this, not only was she surprised, but she was really proud of me for even coming up with the idea, because the implications of this work could potentially affect other women dealing with the same thing,” said Garg.

Garg said that he entered the project with a solid background in coding from taking an AP computer science class in high school and another more basic project that he had worked on previously.

Yet he still faced what he called a “steep learning curve” in OvaScan’s early phases. Garg remembers watching hours of online tutorials as he struggled with aspects of the coding given his lack of experience in machine learning.

He achieved a detection accuracy of around 85% before turning to AI tools like Claude and ChatGPT to help refine the model.

His latest OvaScan prototype achieved detection scores that Garg said were competitive with the best existing models.

Its Dice score, which measures the AI’s predicted tumor boundaries against those drawn by experts, is 0.916, compared with scores of 0.887 to 0.96 from other similar models cited. A Dice score of 1.0 would represent perfect overlap.

Clinical trials would be OvaScan’s next step to test its viability as a tool that can reliably diagnose ovarian cancer in a medical setting.

OvaScan was accepted for presentation at the International Conference on Medical Imaging and Computer-Aided Diagnosis, which will take place in Edinburgh, Scotland from Oct. 22-24.

hhorowitz@dailyherald.com